Sriganesh Srinivasan
← All work

Tech Trends News Agent.

Ask the week's tech news a question

2025 · Team of 2 · CS 4100 (AI) · Archived

Code ↗
Problem
Tech news moves fast and is spread across dozens of outlets, so keeping up means reading a lot of repetitive coverage.
Built
A Streamlit app over live NewsAPI headlines. Questions retrieve the five most relevant articles through a TF-IDF search we wrote from scratch, and a local Qwen2.5-1.5B model answers from them with sources. A second view clusters the coverage into topics by which terms appear together.
Stack
Python, Qwen2.5-1.5B, Hugging Face Transformers, NLTK, Streamlit, Google Colab
Outcome
A working end-to-end agent for our AI course final project. The model loads in about a minute on a Colab T4 GPU, and questions are answered in a few seconds.

How it works

Tech Trends News Agent pipelineTechnology headlines from NewsAPI are tokenized and lemmatized with NLTK and indexed with a TF-IDF search written from scratch. Questions retrieve the top five articles, which a local Qwen2.5-1.5B model turns into an answer with sources. Separately, term co-occurrence clusters the corpus into five topics with sub-trends. A Streamlit app shows both.NEWSAPITech headlines~70–100 per fetchNLTKPreprocesstokenize, lemmatizeFROM SCRATCHTF-IDF indextitles weighted 3×CHAT · COLAB T4Qwen2.5-1.5Banswers from top 5TOPICSCo-occurrence5 topics × 5 sub-trendsSTREAMLITThe appanswerswith sourcestopic charts
Dashed: the external news API. Search, answers and topics all run in the notebook.

Limits

The free NewsAPI tier caps how many articles you get (about 70–100 per fetch), and a 1.5B-parameter model sometimes gives generic answers. A larger model and more sources were the obvious next steps.

Next projectiMessage Wrapped →